Delay-aware evolutionary optimization of digital circuits

Logic optimization, Cartesian Genetic Program-
ming, Evolutionary Resynthesis

In the recent years, machine learning techniques
have successfully been applied in various areas of digital circuit
design including logic synthesis. Evolutionary resynthesis, among
others, represents one of the machine learning approaches. This
technique is based on local iterative optimization of parts of
the original circuit. Even though the local optimization could
be inefficient compared to the optimization conducted on the
whole circuits, it has been shown that the resynthesis performs
extremely well. It produces more compact solutions compared
to the state-of-the art synthesis methods. In addition, it scales
significantly better compared to the evolutionary optimization
performed at the level of the original circuit. The previous methods
have been focused solely on the optimization of the number
of gates. In this paper, we analyse how the local optimization
affects the delay of the resulting circuits and based on that, we
propose a modified approach that considers the delay in the
course of the optimization process. The proposed modification
enables to maintain the delay of the optimized circuit at a
reasonable level without a significant overhead. The evaluation
conducted on a set of non-trivial highly optimized benchmark
circuits representing various real-world circuits demonstrated
that the proposed method is able to remove a significant number
of gates while preserving the delay within the requested bounds.